When Midjourney first took off in the summer of 2022, I joined their discord and quickly used up my free generations. Then I started looking around for alternatives. Everything else that was available (e.g., DALL·E) was disappointing by comparison. So my interest faded away. But then in January of 2023 I discovered a slew of new apps, based on Stable Diffusion, and started trying them out. My interest was renewed.
So when Adobe announced beta access for their own AI, called Firefly, I applied for access and got it. The results were somewhat disappointing. Beautiful art is possible, for sure; but I had already honed my skills at writing prompts for Stable Diffusion and those prompts didn't do well in Firefly. Rather than learn how to write good prompts for Firefly, I decided to see if I could get Stable Diffusion to produce something good using one of Adobe Firefly's public prompts:
a fuzzy psychedelic caterpillar, baroque
Here's what that prompt gave me using Adobe Firefly:
The first one (top, left) is the example from Adobe; the other four I generated. They're interesting, but they come with an ugly watermark, which Adobe insists is for ethical purposes.
Here's what I got from Stable Diffusion 2.1 using the same prompt:
To paraphrase Vitruvius from The Lego Movie, "That is just…the worst."
So then I ran those five images from Adobe Firefly through CLIP Interrogator 2.4. Here is the prompt that was generated for the first image:
there is a toy bug that is sitting on a mushroom, Zbrush contest winner, the caterpillar, living spore microorganisms, portrait of a cute monster, by Alexander Kucharsky, portrait of a cartoon animal, pollen, centipede, colorful closeup shot, Jean-Sebastien Rossbach, by Zlatyu Boyadzhiev, microscopic
And here is what I got from Stable Diffusion 2.1 for that prompt:
Those don't look like caterpillars, so I modified the prompt a little:
there is a caterpillar, teal orange mauve, red spots, Zbrush contest winner, the caterpillar, living spore microorganisms, portrait of a cute monster, by Alexander Kucharsky, portrait of a cartoon animal, pollen, centipede, colorful closeup shot, Jean-Sebastien Rossbach, by Zlatyu Boyadzhiev, microscopic
And here is what I got from Stable Diffusion 2.1 for the modified prompt:
Here is the prompt that was generated for the second image:
there is a very colorful snail with many small balls on it, Sylvain Sarrailh and Igor Morski, RNA bioweapon, colossal fluffy tardigrade, CG Society contest-winner, in style of Beeple, the caterpillar, stitches, closeup of an adorable, C4D style, colorful
And here is what I got from Stable Diffusion 2.1 for that prompt:
Those don't look like caterpillars, either. But I didn't feel like playing around with the prompt anymore.
Here is the prompt that was generated for the third image:
there are two fuzzy caterpillars on a green plant with yellow and blue leaves, giant centipede, Behance, high detail, electron microscopy photography, CG artist, closeup of an adorable, buggy, a colorful, his trunk is a long tentacle, inspired by Mike "Beeple" Winkelmann, highly detailed 4K art, Behance art
And here is what I got from Stable Diffusion 2.1 for that prompt:
Some of those are pretty good.
Here is the prompt that was generated for the fourth image:
there is a very cute caterpillar that is sitting on a mushroom, mandelbulb 3D fractal, the caterpillar, cyberpunk pincushion lens effect, Artem Demura Beeple, portrait of a cartoon animal, realistic highly detailed, hyperdetailed colorful, extremely detailed, cubensis, bug
And here is what I got from Stable Diffusion 2.1 for that prompt:
Looks like Stable Diffusion had trouble making just the caterpillar fuzzy.
Here is the prompt that was generated for the final image:
there is a small stuffed animal that is sitting on a mushroom, in the art style of Filip Hodas, centipede, colorful high contrast HD, pollen, portrait of a cartoon animal, wallpaper mobile, the caterpillar, Andrey Gordeev, microscopic picture, highly detailed, buggy, fluffy
And here is what I got from Stable Diffusion 2.1 for that prompt:
Once again, the mushroom came out looking fuzzy, too.
As you can see, the way that Adobe Firefly handles prompts differs from the way that Stable Diffusion 2.1 handles prompts. I like some of the images that I got from Stable Diffusion, but none of them really match the images produced by Adobe Firefly.
Prompts were generated using CLIP Interrogator 2.4.
Illustrations were drawn using Adobe Firefly (beta) and Stable Diffusion 2.1.
